Human fascination with whale song helped turn Songs of the Humpback Whale into a striking cultural success. The vinyl reached 62 on the charts, appeared in National Geographic, and even traveled into space on Voyager.

The appeal is not difficult to understand. Haunting clicks, whistles, and whines travel through the water, creating captivating soundscapes even though humans still do not know what the whales are saying.

Now, researchers are pursuing an idea that once sounded safely confined to science fiction: translating those sounds and potentially sending messages back. The effort could eventually put unsuspecting whales in conversation with humans and their machines.

Project CETI, short for Cetacean Translation Initiative, is leading the attempt at interspecies communication. According to Hakai Magazine, the project began its ambitious work in March 2020.

Its goal is to decode whale songs, determine whether whales have what humans would recognize as “language,” and hopefully find a way to talk back. Such an exchange would turn years of human eavesdropping into something closer to an actual conversation.

The idea emerged through a series of chance discussions among researchers. Computer scientist Shafi Goldwasser and marine biologist David Gruber began by considering the similarities between sperm whale clicks and Morse code.

Goldwasser and Gruber later joined forces with computer scientist Michael Bronstein. He proposed using AI to examine enormous collections of sperm whale recordings and identify patterns that might be comparable to speech.

That approach raises a difficult question about when communication becomes language and whether language truly exists outside humans. Whale sounds may contain patterns, but determining what those patterns mean presents a much larger challenge.

Dr Valerie Vergara has spent much of her career listening to highly vocal beluga whales. In a recent interview with IFLScience, she discussed cetacean communication and the complex behavior behind the animals’ many sounds.

Belugas are known as the “canaries of the seas” because of their noisy nature. Young whales display their own form of beluga babble as they attempt to learn vocalizations from their parents and the wider pod.

One emerging part of Vergara’s work examines whether individual beluga whales have unique vocal signatures. Decoding those signatures could be pivotal in determining whether belugas talk and whether their exchanges qualify as conversation.

Turning cetacean sounds into material researchers can analyze requires processing a vast amount of data. Human researchers cannot handle that volume alone, which is where AI becomes central to the project.

Language models such as GPT-3 learn what conventionally comes next in a sequence and can complete an unfinished sentence or headline. The process resembles autocorrect, although GPT-3 can still produce catastrophically wrong results.

The scale of the whale data problem is formidable. Technology used for human language has required roughly 175 billion words, while the current collection of sperm whale “codas,” the term for the sperm whale equivalent of a word, contains about 100,000 examples.

Project CETI therefore needs to expand the number of sperm whale recordings substantially before an AI neural network can be adequately trained. Even if that succeeds, nobody knows how whales might receive the resulting messages, and Bronstein offered one memorable possibility: “Maybe they would just reply, ‘Stop talking such garbage!’”